The argument in one line.
The only thing stopping non-technical people from running one of the most capable open-source AI agents was a terminal setup wall, and the new desktop app just removed it entirely.
Read if. Skip if.
READ IF YOU ARE…
- You tried to set up a local AI agent before and quit the moment you saw the terminal.
- You want an AI agent running background tasks on your machine without paying a monthly SaaS fee.
- You use Telegram or Discord and want to talk to a local agent from your phone.
- You are evaluating open-source agents against hosted solutions for everyday automation.
SKIP IF…
- You need 24/7 uninterrupted scheduled automation -- the agent sleeps when your computer does.
- You are already comfortable running Hermes via the terminal and have nothing to gain from a GUI wrapper.
The full version, fast.
Hermes is a self-improving open-source agent with 183K GitHub stars that runs locally, maintains persistent memory, and writes its own skill playbooks over time. Until now it required terminal fluency, which blocked most non-technical users. The new desktop app from Noose Research installs the full stack automatically. The interface has four sections: Skills and Tools (109 defaults to trim down), Messaging (pipe the agent into Telegram, Discord, Slack, or email), Artifacts (file output storage), and Cron (schedule plain-English jobs). The hard constraint: scheduled jobs only fire while your machine is awake, so for always-on automation you need a dedicated computer or a VPS.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 01:03
01 · Hook and framing
Hermes positioned as the most powerful available agent; terminal friction named as the adoption killer
01:03 – 02:04
02 · What Hermes actually is
Noose Research repo shown, 183K GitHub stars, self-improving memory and skill-writing loop explained
02:04 – 02:17
03 · Why the desktop app matters
The terminal wall framed as the single biggest friction, now eliminated
02:17 – 03:15
04 · Download and install
Website download, one-click installer, automated Python venv creation -- no terminal needed
03:15 – 04:05
05 · Free and open source -- you just need a model
App is free; requires LLM API key from OpenAI, Anthropic, OpenRouter, or Noose subscription
04:05 – 05:41
06 · Skills and tools: trim the fat
109 default skills shown; recommendation to disable unused ones to cut token cost and latency
05:41 – 07:32
07 · Messaging: talk to it from anywhere
Messaging section with Discord, Telegram, Slack, WeChat, Twilio, email -- connect the agent to any channel
07:32 – 09:45
08 · Cron: scheduled jobs
Cron section shown; existing weekly X profile growth review present; model-switching from status bar
09:45 – 11:09
09 · Build 1: countdown timer
Live build -- Hermes asked to create a dark-background HTML countdown timer; output shown in artifacts
11:09 – 12:03
10 · Build 2: morning news brief
Scheduled job in plain English -- top 3 AI news stories every weekday at 7AM to Telegram; scheduled in 28ms
12:03 – 13:05
11 · The catch: computer must stay on
Scheduled jobs sleep with the machine; Mac Mini always-on or VPS as workarounds
13:05 – 14:31
12 · Honest caveats and CTA
Early public preview status acknowledged; community and Reprise business pitch
Lines worth screenshotting.
- The terminal was never the agent itself -- it was the setup tax that eliminated most potential users before they ever reached the value.
- 109 skills are enabled by default in Hermes; each active skill adds token overhead to every request, meaning most installs are silently costing more than they should.
- You can schedule a recurring AI task in plain English and have it confirmed in 28 milliseconds -- no cron syntax, no configuration file.
- Connecting an agent to Telegram costs nothing extra and lets you issue tasks from your phone with zero additional tooling.
- A desktop app that mirrors a terminal setup is not a simplification -- it is a different product with a different user base.
- Self-hosted agents have a hard ceiling: the machine must stay on for scheduled jobs to fire, which is the one constraint SaaS avoids by default.
- Open-source star counts are a meaningful signal for agent projects -- 183K stars puts Hermes in the same tier as established developer tools.
- The artifact section of an agent UI solves a real friction point: output files accumulating across chat sessions are impossible to find without a dedicated collector.
The wall between most people and local AI was never the technology.
WHAT TO LEARN
Hermes has been capable for a long time -- what changed is that the setup friction is gone, which is a reminder that distribution is often the actual bottleneck in technology adoption.
- The terminal is a hard filter that eliminates the majority of potential users before they ever experience the product -- removing it is not a feature, it is a different product.
- Default-on settings in AI tools carry hidden costs: 109 active skills in Hermes each add token overhead, meaning an unconfigured install is measurably more expensive than a trimmed one.
- Scheduling a task in plain English and receiving confirmation in 28 milliseconds is a useful baseline for evaluating what counts as a genuinely simple interface versus what just looks simple.
- Self-hosted agents have a hard constraint that hosted services do not: the machine must stay on, so understanding which constraint applies to your situation should happen before choosing the tool.
- Connecting an AI agent to an existing communication channel like Telegram lets you interact with it from any device without building a new interface -- meet the tool where you already are.
Terms worth knowing.
- Hermes
- An open-source self-improving AI agent built by Noose Research that runs locally, maintains persistent memory, and learns workflow shortcuts by writing its own skill files.
- Noose Research
- The research lab that built and maintains the Hermes agent project and its desktop application.
- Skills
- Pre-built capability packages inside Hermes covering areas like file operations, web search, and social media. Each active skill adds token overhead to every prompt.
- Cron
- The Hermes scheduler (labeled Chrome in the app UI) that lets you define recurring jobs in plain English. Jobs only execute when the host computer is awake.
- OpenRouter
- A routing service that gives a single API key access to hundreds of different language models from multiple providers.
- VPS
- Virtual Private Server -- a rented remote computer that stays on continuously, used here as a workaround for running Hermes 24/7 without leaving a personal machine running.
Things they pointed at.
12:09productHostinger VPS
12:14productHetzner VPS
Lines you could clip.
00:00
“Hermes is probably the most powerful AI agent that you could be running right now.”
01:30
“All of that power lived behind the terminal. And that is the wall that stopped most people completely cold.”
04:30
“Every skill that is on, it is a little extra weight that the agent actually carries around, and that can quietly cost you tokens -- which is real money.”
11:38
“In just twenty eight milliseconds -- done, it is scheduled.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
00:00Hermes is probably the most powerful AI agent that you could be running right now. It works for you all day. It's got its own memory, and it writes its own skills the more that you use it.
00:08But up until this week, just to set it up, you had to go into the terminal. And if you're not a technical person, you take one look and you quit. Now I've been running Hermes for months now, and the agent was never the hard part.
00:18Just getting everything installed and getting my actual setup to the way that I want it to be was the hard part. So I've watched a ton of different people inside of my community just give up before they ever got to the good stuff. So Noose just dropped a desktop application to solve for that, where you can just download it like any other application.
00:34It does the whole setup for you, and a couple minutes later, you're talking to it like a window that looks like this. There's no terminal needed to set this up. No code.
00:42None of that. So I'm gonna be showing you guys four different things, just what Hermes actually is and why this application is such a big deal, how the install just runs itself, how simple it actually is, and a quick walk through of the application, and then we'll connect it to Telegram and get it working for you. And look, if you can install Slack or any other simple application onto your computer, you can do this.
01:02I promise. Let's get into it. Now really quick before we install anything at all, let me just explain what this thing actually is because it's exactly why this application matters so much.
01:10This is the Hermes repo. It's built by a research lab called Noose Research. It's open source, and the star count at the top, a 183,000, is actually pretty wild.
01:18So for an open source agent, that is pretty massive. It's up there with OpenClaw, And their tagline, it pretty much sums it down. Their tagline down to the bottom, it pretty much sums up what this is.
01:28The agent that grows with you, Hermes, is in the chat bot that you open a tab and ask a question. It runs quietly in the background on your computer with its own memory, and every time it does a task for you, it writes a little note on how it actually pulled it off. So next time, it's gonna be faster, and it just gets better at being yours over time.
01:45Now the catch with this, and it's the whole reason for this video, all of that power, everything that you're looking at here up until now, it lived behind the terminal. So the commands, the setup wizard, the wiring things up by hand, and that's the and the biggest friction. So that's exactly what this desktop application fixes.
02:04To go ahead and head to this link, either just pause on the video or you can grab it from the link down below in the description, and we're just going to scroll down a little bit. You can see we could download this for Mac OS.
02:14It already detected that that is the computer that I am using. It's actually one of the best looking website landing pages I have ever seen, so let's just do that now. So whenever you go through, you download this on whatever respective machine you are using, it'll pull up once you actually open this up.
02:28Hermes agent, obviously, I wanna click on install, and it's going to handle everything for us. So it's actually kind of magic. It's just setting everything up.
02:36So it's pulling down the agent. It's building the little environment that needs to run on your machine, and then just getting everything ready to go. Basically, all the stuff that you used to have to type in the terminal by hand, it just does all of it for you.
02:49So some of the things, downloading Hermes agent, creating the Python virtual environment, installing the browser tool dependencies, installing the Hermes command, everything.
02:57Couple minutes later, everything's now set up, and it says Hermes is ready, we can launch Hermes. From here, just gonna continue the setup, make sure everything's good to go, start up Hermes for us. Fantastic.
03:06So once that is finished running, another five seconds or so, we will now have this interface. Everything loaded. It is strikingly similar to Codecs or really any other IDE out there where we have the sessions up the top, And then we have this navigation bar on the right hand section.
03:22We can see all of our folders and everything that we can work inside of. On the left hand side, we can see all of our sessions, some pinned things that we want to throw in there, anything that may be important, artifacts, messaging, skills and tools, and starting a new session, of course.
03:34Now this app itself, it's completely free, and it's open source. You just need a model, so the brain behind it. So I am going to be using ChatGPT for mine.
03:42Noose also has their own subscription just with over about 300 different models. So if you want the easy all in one way, or you can just plug it into Anthropic or whatever else, can use that API or OpenRouter is gonna be great options. All of those work completely fine.
03:57Also, Noose, they do give you a little bit of credits to try out, but that's really just kicking the tires. So to really use it day to day, you're gonna be on a cheap plan or just use your own key. And one thing before moving on that I did want to show off a little bit is the skills and tools.
04:10So out of the box, Hermes comes loaded with a stack of different skills and different tools, and I mean a lot of them. So mine's showing about a 109 different skills right now, and all of these are sorted into different categories.
04:21We have Apple, autonomous AI agents, creative, data science, everything, all the way up until social media, note taking, media, I mean, you name it.
04:30One thing that most people miss is that if you see how almost all of these are just switched on by default. So every skill that is on, it is , and that can quietly cost you tokens, which is real money. So I just go down the list, and I should be switching off the stuff that I'm never going to be touching.
04:48So let me go ahead and do that. Now right next to this skills tab is going to be the tool sets. So these are the actual abilities all grouped up.
04:57So anything like file operations, so that it can actually, you know, like, read and write files. Vision, so it can just look at images and memory and test delegation, it can spit up just little helper agents to work in parallel.
05:10So it's the same thing. You can flip them on and off, whichever ones you do not need. Highly recommend first thing to just switch off whatever you're not going to be using because, again, it'll make it a little bit more expensive and just take a little bit longer for you to be running any outputs.
05:22And by the way, a few of them do say that they need keys, which just means that you would plug in a key to switch that one on and off. And one more thing that I love is this is also where the skills that it writes for itself is going to be showing up inside of this section. So the little playbooks, it's going to be saving as it learns how you actually work moving forward.
05:41Next piece is messaging, and this is the one that kind of got me, because this is where you hook Hermes up to the chat applications, and you can talk to it from anywhere. It is so extremely simple, I cannot even fathom it. The list that they actually have of everything in here, it's actually pretty long and very extensive.
05:56So Discord, Telegram, Slack, Mettermost. I haven't even heard of a majority of these. They have Twilio, email.
06:02So I mean, we can text and email any of our customers or any prospects. A lot of these I haven't even heard of. WeChat is a big popular one, Google Chat, you name it.
06:12So effectively, wherever you currently talk right now, you can drop your agent in there. So you can just pick one of these, gives you a little setup guide right here, so we can go to literally any one of them. We'll find any setup guide for each respective channel, and be able to follow that, be able to paste in our tokens, whatever it may be, and connect.
06:28And we're going to be doing Telegram specifically together in just a minute, so you'll see exactly how it works. Next up from there is going to be the artifacts. Now the artifacts, it's effectively the home for everything that comes out of your sessions.
06:39So the files that it builds, any images, or anything that you have, it is going to be saving. So the files that it builds, any images, anything that you have it save, it all collects it just right here, so you're not digging back through an old chat trying to find that one thing, whatever it may be once again. Now below this, if you go down to the bottom left hand section, you can see we have this status bar.
07:02So gateway, that's just going to be your your messaging connection. We'll come back to it, and then we have the agents right next to that. Of course, we don't have any live sub agents right here, right now.
07:13But this is just going to be showing any helper agents running right now. And then if we close out of this, we then have the Chrome. So this is just the handy one, where anytime you set up a job to run on a schedule, it lands right here in the Chrome where you can see, you know, anything.
07:27You can pause it, test it, or click in, and you can edit any one of these. We can create on a new one. We currently have a weekly x profile growth review.
07:35Now funny thing is is I set up this months ago where I would set up my Hermes agent through the terminal, and it's now recognizing that I already have everything set up. So it's just taking all of that. So if you do have a previous terminal set up with everything, there's not really much you have to do as long as you go through the entire setup that I just showed you guys.
07:54But anyways, if you back out of that, move all the way to the bottom right, we can see the model that we are currently running. This makes things so much easier because previously, I would have to go back in the terminal, see all of the different models that I had by opening up my config, opening up the settings, and it it just I didn't memorize all the different commands and things that I had to paste in.
08:12I'd always have to speak with another AI system, another AI model to try to figure it out. But everything here, it's now concise, and I can literally just go back and forth because I already have my open router set up, and I have my OpenAI ChatGPT subscription plugged in here as well.
08:26So I can just easily navigate back and forth through each one. Extremely simple. So much better now.
08:31Outside of that, we have our settings and the settings in here. It's effectively just going to be like the control room for the entire thing. So I'm not gonna walk through every single tab, but there are a few that are worth knowing.
08:43So this is just where, of course, you manage your model and your providers. So the account or the key that's actually going to be powering it, you plug it in here. Next up is going to be the appearance.
08:55So this is just going to be your themes. And there's a memory and a context tab, which is the one that I would actually be be pointing you to. So this section right here.
09:03If it ever feels like your agent is forgetting some things, this is where you tune how it remembers. So you can just tweak these and nudge these a little bit, and it's going to hold on to stuff way better for you depending on what you change specifically. And one of the last things down in this about section right here, is this just where you can see whatever version that you are on, and the nice part is it checks for updates on its own, and it tells you when there's a new one, so you're never going to be stuck on something old.
09:25Okay. Like I mentioned, we're going to be using Telegram. We can see instead of the messaging, we are already connected, how you can actually set this up for yourself.
09:32Of course, you can just open up the setup guide, but you're going to grab your bot token from Telegram. Simply just open up Telegram, create a new contact, and search for bot father, and it's going to present you with a code once you enter in the name that you want to name your bot. Simple as that.
09:45So once you get that set up, we are going to get into the actual builds, and I wanna show you how to do something with this. So we're just gonna create a brand new session and ask it to just build me something little. So what I'm going to ask is build me a simple one page countdown timer to a date that I give you.
09:58Clean look, dark background, big numbers, save the file when you are done. So right now, you, of course, just like any other AI model like Claude or maybe like GIGPT, it is just going to be thinking it through, picking its tools, or writing the thing, and then running it.
10:11And this is the part that I really like about the new app is is you can finally see what it's doing instead of just staring at a blank chat, which was super annoying with just using it inside of Telegram. So from now on, I'm only going to use Telegram when I actually have to just use my phone, and I'm not at my desk. Now anyways, it's asking what date should we do the countdown.
10:28So let's just do New Year's Eve. Beyond that, if we click on any one of these little tabs, we can expand on any one of these, and we can see what's actually going on inside of that beyond what I was just talking about.
10:40And just like that, we now have our output. We can just click on the previews inside of here. Super easy to navigate.
10:45There's our HTML. And it looks like okay. No.
10:48There's both two outputs, but we just got one. So exactly what we are looking for. Beyond that, if we go inside of our artifact section, we should be able to see it inside of here.
10:56So if I just filter I don't know why all of these ones are coming up specifically because none of these are mine. It looks like it's some sort of community. But if we just scroll down here, we can see our HTML right here and all of our other previous ones.
11:09Beautiful. Now what I wanna build is something that runs every single morning on its own, and this is where it stops being a simple chat tool and turns into something that's actually going to be working for you while you are asleep. And the wild part is that you just set it up by just asking.
11:20So I'm gonna say is every weekday at 7AM, grab the top three AI news stories from the last twenty four hours and send me a short summary of each one. So Hermes, what it's going to do from right here, it's going to think, run its Chrome tool, and reply with just a results card laying out the schedule, the next run, and that is just going to be repeating forever.
11:36So just like that, in just twenty eight milliseconds, done, it's scheduled, here's the job, here's the schedule, the next run, the delivery, and even a job ID. So with that, I built a scheduled job.
11:44We can open up our skills and tools. It's like that if we open up our going down to the bottom left, we should see it pop up here right now. We can also open this up.
11:54We can trigger it now, edit it, the frequency, who to deliver this to, whether we wanna deliver it to Telegram, Discord, Slack, email, and when we want this to be running down at the bottom. The one thing that I do have to be honest about this is this runs on your computer. So that job only fires when your computer is on.
12:09So if you do shut your laptop, it is going to sleep, and the agent sleeps right with it. So if you want this every morning no matter what, you have to put it on a machine that stays on. For example, I have another Mac Mini in my other room, which is just always on every single day, in which case it's running Hermes for me.
12:25Or if you just want to run this on a VPS, you can use something like Hostinger, or you can use Hertzner. There's a lot of different options, in which case you can put Hermes on a VPS, but there's pros and cons to each one of those.
12:36But we have covered that in previous videos. Now another great thing about this is the cross functionality. So if I'm in Telegram, maybe on my phone, I can just type out, I need you to say hello back to me.
12:48So we type this out. We're gonna get our hello back, and if we just go back into our Hermes application, we should see that's that session in just a second if we actually refresh it here.
12:59So we have the simple greeting. Perfect. Now everything is just translated back in here, so we now have a cross functionality.
13:05So that's everything that I wanted to cover, and I want to be straight with you about where this is still a little rough because it is brand new, so it's an early public preview. So pretty much all of this, it's moving fast and things will change, and the two catches that I already mentioned still stand. But the free credit is just a trial, and your scheduled jobs only run while your computer is on.
13:22And if you want something for heavy serious coding all day, this isn't just trying to be that. Here's the point instead. Hermes has been one of the most capable agents out there for a while now, and the only thing holding it back was that you basically had to be a terminal person just to switch it on, and that wall is now completely gone.
13:37You just download an app, click through it, and you're running it. And that's a genuinely big deal, and it's exactly why I wanted to get this in front of you. So if you go set this up, drop a comment, tell me which one you built first, and the little application, or morning brief.
13:51If you guys are gonna be using this, setting this up for yourself, having it be kinda like your main harness, let me know down in the comments. I'm very curious. Also, make sure to check out our free school community.
13:59We have so many in-depth guides, so many opportunities, and we hope you grow and start your AI business. There's 19,000 other people inside of there.
14:08You can meet people for partnerships. You can find hiring opportunities. There's so much good stuff and so much more to come, monthly competition.
14:15So if you guys are interested, make sure to check out the link down below in the description. And, course, if you're looking to grow your business in 2026, then book in a call with our team at Reprise. We'll show you how we can drive leverage and increase your bottom line inside of your business.
14:27So make sure to check that out. Book in a call with our team, and I'll see you in the next video.
The Hook
The bait, then the rug-pull.
Hermes has 183,000 GitHub stars and a self-improving memory loop that gets better the more you use it. For months, the only thing between most people and that power was a terminal prompt they never opened.
Named ideas worth stealing.
00:34list
Four sections of the Hermes desktop app
- Skills and Tools
- Messaging
- Artifacts
- Cron
The app organizes everything into four panels: capability management, external channel connections, output storage, and scheduled job management.
Steal forstructuring a product walkthrough for any multi-section tool
How they asked for the click.
VERBAL ASK
13:05product
“If you guys are gonna be using this, let me know down in the comments. Also make sure to check out our free school community. And if you are looking to grow your business in 2026, book in a call with our team at Reprise.”
Community plug first, then hard agency CTA. Two distinct asks in rapid succession with no visual separation.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
